AI Consultant · Ireland
Mercurious.ai is an independent AI consultancy in Ireland for professional services firms. I find the repetitive, non-billable work eating your best people’s weeks, count it, and help you decide what should be automated, assisted, fixed, or kept human. The technology comes last, and sometimes the answer is that you don't need any AI.
The problem
A professional services firm sells judgement, expertise, and the relationships built on both. That is what the client is paying for, and it’s often the only input listed on the invoice.
Then look at where an experienced person’s week actually goes. Chasing. Preparing. Assembling. Reporting. Searching for something somebody filed in 2023. Updating a record. Drafting the same message for the fortieth time. Writing up what was already discussed and agreed.
That work never appears on an invoice, and nobody has ever added up how much of the week it takes.
Take that work off them and you get their week back for the work clients actually pay for. That is the whole proposition, and everything below is how I go about it.
Category
It is a job title with very little agreement behind it, so it is worth saying what I think it should mean.
A useful AI consultant does not open by recommending ChatGPT, Copilot, a platform, or an agent. Those may all turn out to be part of the answer. None of them is a diagnosis.
What should come first is an account of the work:
Only then is anybody in a position to say what technology belongs in the answer.
How to find the work AI should actually do →
The person behind this
I’m Rob Abbey, a product manager and an AI practitioner rather than a developer. I spent a decade building software for US technology companies, the kind with real users, real revenue, and real consequences when something broke: six years at ClickFunnels as the user base grew from 6,500 to over 100,000, then Principal AI Product Manager at a company with $10M+ ARR, shipping chatbots, voicebots, prediction models, and a migration from NLP to LLMs. The last five years have been AI in production.
More about how I work →The work
What comes up in nearly every firm:
The boundary
The approach
I have a two-stage approach to AI automation. First, I find out what we need to build, and then I build it incrementally.
One facilitated day inside your firm, ideally with people from all relevant departments in the room, because everyone has very different tasks and time sinks.
No technology is chosen in advance. We just focus on the jobs to be done.
How a Census day runs →Where the Census says work should be assisted or automated, we either build or buy a solution, or a combination of both. For a build, I design and deploy a small team of AI agents around it: briefing, chasing, reporting, drafting, keeping what the firm knows, and minding the whole thing.
All of the above is managed, not just installed. Software requires oversight, and that’s exactly what I provide. Finally, when an agent stops earning its keep, I say so, and we agree to retire it.
About the Fleet →In practice
Accountancy practices are the clearest example I know of. Hundreds of clients need the same records, the same reminders, and the same recurring work at the same point in a calendar that Revenue and the CRO write rather than the firm.
No single one of those jobs is heavy. The repetition is what makes them heavy, and it is why an hour of a qualified person’s week disappears without anyone being able to point at where it went.
AI automation for accountancy firms in Ireland: what to automate first →
Fit
Irish professional services firms where:
In practice: accountancy practices, advisory and consulting firms, solicitors and other professional practices, and specialist expertise-led businesses that run on documents.
I’d point you elsewhere if:
The principle
Because a tool-first process ends, more often than not, with capable software going round a business looking for a job worth doing.
Start from the work and the answer is allowed to be any of these:
Before you automate anything: how to find the work AI should actually do →
Working together
It’s a straightforward services ladder: the Orientation, then the Census with its AI Roadmap, then deployment if it’s warranted. Each step is priced on its own and funds the decision about the next, and you can stop at any rung.
And if you’d rather start on your own, the free Repetitive Work Calculator puts numbers on one repetitive task and runs it through the five-way sort. No booking involved.
One hour, online. You tell me where the work is piling up and what you’ve already tried. I tell you whether AI helps here and what I’d do first if the firm were mine. Written recommendations within 48 hours, yours to keep. Inside the Orientation →
One facilitated day. Find the work, count it, weigh it, and run every job through the five-way sort. Within 48 hours you get the AI Roadmap: what was counted, what it weighs, what I’d hand over first and why, and what I’d leave well alone, with the reasoning shown so you can argue with it. Ninety days to decide, the price holds for those ninety days, and the whole €1,850 comes off a deployment if you go ahead. How a Census day runs →
The fleet built and deployed around your firm and the way you actually work, on your own accounts and keys. About the Fleet →
Watched daily, improved monthly, one page in your inbox at month end, and a Quarterly Count where you pick what joins next. It is most of what having a fleet actually is.
Software needs to be taken care of. Models change, APIs crash, keys expire, and prompts drift. Someone has to take care of what you’ve built, and the best reassurance is being able to say “Rob looks after that for us”.
And for firms that want one person to take the AI question off their plate and take responsibility for it, a day a week: fractional AI leadership →
Common questions
A useful one starts with your work rather than with a product. Before anything is recommended, the questions are what work exists, how often it happens, how long it takes, how much judgement it needs, whether the source information is actually there, and what happens when it goes wrong. Only after that does it make sense to talk about what technology belongs in the answer, if any does.
An automation agency is set up to build automations. My process starts a step earlier: the Census counts the work first, and a fair share of what it finds is better fixed than automated, or better left with a person. The five-way sort allows every one of those answers.
The other difference is what happens after the build. The Fleet is managed month by month, and when an agent stops earning its keep I say so and we retire it.
No, and it is generally better if you haven’t decided. Arriving with a tool already chosen tends to shape the assessment around justifying it. Bring the work and the frustrations instead.
Almost exclusively. The firms I do my best work with sell time and expertise, produce a lot of documentation, and carry a lot of repetitive administration around the client work. That is usually somewhere between five and fifty people for the Census and Fleet route. From fifteen up to about seventy-five, the fractional arrangement, a head of AI one day a week, tends to be the better fit. Well above that, you have the scale to hire someone internal, and I would tell you so.
Preparation, mostly. Meeting briefs, routine follow-ups, chasing outstanding information, recurring reporting, document assembly, practice-system updates, and finding what the firm already knows. The common thread is that the work repeats, follows a method, and can be checked quickly by a person.
Not in anything I build. Drafts are prepared and queued for a person to approve. Nothing consequential is to be done without somebody authorising it. That rule goes in before an agent is given access to anything that can send.
Professional advice, material judgement, final sign-off, difficult client conversations, and anything written about a person. I would also keep a human on any job that fails quietly, where a mistake would go unnoticed for months. Repetition alone is not a reason to hand something over.
I am based in Athlone, in the midlands, and I travel for Census days anywhere in Ireland. Orientations run online. The date is set around your team rather than my diary.
The AI Orientation is €450 for an hour, with written recommendations within 48 hours that are yours to keep. The Census is €1,850 for a full facilitated day, and the whole of it comes off a deployment if you go ahead within ninety days.
Fleet pricing is settled after the Census, once there is something real to price.
A written AI Roadmap lands within 48 hours: what was counted, what it weighs, what I would hand over first and why, and what I would leave alone. You have 90 days to decide, and the credit price holds for the same 90 days.
If the answer is that AI is not worth your time yet, that is in the AI Roadmap too, with the reason and what would have to change.
If you’re deciding where to start
Longer pieces, written for the questions firms actually arrive with.
If experienced people in your firm are spending too much of the week preparing, chasing, assembling, reporting, and writing things up, the Census is where I’d start. One day to find the work, count it, and decide what deserves changing, before anybody buys or builds a thing.